AI agents

Test an AI agent and its human handoff

Check answers, limits and handoff behavior before an AI agent handles customer conversations.

Published · Updated

The agent test panel lets you review answers and actions before customers see an unfinished experience. Test success, uncertainty and human handoff—not only an easy question.

Before you start

Prepare an agent with initial instructions and knowledge, and identify the team that should receive cases requiring a person.

Steps

  1. 1

    Open the test panel

    From the agent creation or edit page, open testing and send a realistic customer question.

  2. 2

    Test missing knowledge

    Ask about something absent from the sources. Confirm that the agent does not invent an answer or make an unapproved commitment.

  3. 3

    Request a person

    Send a clear request to speak to a teammate or a case that requires human intervention.

  4. 4

    Review the outcome and context

    Check that the agent chose the handoff path and that the team receives available conversation context as configured.

  5. 5

    Improve and retest

    Correct instructions or actions when behavior is wrong and repeat the scenario before publishing. Review run logs after activation.

Check the result

The agent stops or hands off when expected, and a teammate can continue from the available context without making the customer repeat everything.